Major PeaceBuilders Research Program Articles

Major PeaceBuilders Research and Program Articles

The following publications are attached, which provide
information and background on the large-scale, community-based
randomized control trial evaluating a theoretically-driven violence-
prevention program, PeaceBuilders®.

References Cited

Embry, D. D., Flannery, D. J., Vazsonyi, A. T., Powell, K. E., & Atha, H. (1996).
PeaceBuilders: A theoretically driven, school-based model for early violence prevention.
American Journal of Preventive Medicine, 12(5, Suppl), 91.

This paper describes the theory, experimental design and baseline findings of the
randomized control group study.

Krug, E. G., Brener, N. D., Dahlberg, L. L., Ryan, G. W., & Powell, K. E. (1997). The
impact of an elementary school-based violence prevention program on visits to the school
nurse. American Journal of Preventive Medicine, 13(6), 459-463.

This paper reports on reductions in illnesses and injuries as measured by nurses’
office visits. This is possibly the first experimental proof of actual proximal
reductions in violent injuries as a result of a violent-injury prevention effort.

Flannery, D. J., Vazsonyi, A. T., Liau, A. K., Guo, S., Powell, K. E., Atha, H., et al.
(2003). Initial behavior outcomes for the PeaceBuilders universal school-based violence
prevention program. Developmental Psychology, 39(2), 292-308.

The paper provides teacher and self-report outcomes using standardized
measures related to social competence and aggression. Both measures have
strong prediction for life-course aggression or resiliency.

Vazsonyi, A. T., Belliston, L. M., & Flannery, D. J. (2004). Evaluation of a School-
Based, Universal Violence Prevention Program: Low-, Medium-, and High-Risk
Children. Youth Violence and Juvenile Justice, 2(2), 185-206.

This paper reports on differential effects of the intervention for the most-high risk
youth, based on baseline data.

The People Magazine article from April 5, 1999 tells the story of the use of
PeaceBuilders in Salinas, CA and other communities.

Developmental Psychology Copyright 2003 by the American Psychological Association, Inc.
2003, Vol. 39, No. 2, 292–308 0012-1649/03/$12.00 DOI: 10.1037/0012-1649.39.2.292

Initial Behavior Outcomes for the PeaceBuilders Universal School-Based
Violence Prevention Program

Daniel J. Flannery Alexander T. Vazsonyi
Kent State University Auburn University

Albert K. Liau Shenyang Guo
Kent State University University of North Carolina at Chapel Hill

Kenneth E. Powell Henry Atha
Centers for Disease Control and Prevention Pima County (AZ) Community Services Department

Wendy Vesterdal Dennis Embry
University of Arizona PAXIS Institute

PeaceBuilders is a universal, elementary-school-based violence prevention program that attempts to alter
the climate of a school by teaching students and staff simple rules and activities aimed at improving child
social competence and reducing aggressive behavior. Eight matched schools (N 4,000 students in
Grades K–5) were randomly assigned to either immediate postbaseline intervention (PBI) or to a delayed
intervention 1 year later (PBD). Hierarchical linear modeling was used to analyze results from assess-
ments in the fall and spring of 2 consecutive school years. In Year 1, significant gains in teacher-rated
social competence for students in Grades K–2, in child self-reported peace-building behavior in Grades
K–5, and reductions in aggressive behavior in Grades 3–5 were found for PBI but not PBD schools.
Differential effects in Year 1 were also observed for aggression and prosocial behavior. Most effects were
maintained in Year 2 for PBI schools, including increases in child prosocial behavior in Grades K–2.
Implications for early universal school-based prevention and challenges related to evaluating large-scale
prevention trials are discussed.

Despite recent downturns in national rates of violence perpetra- significantly high levels (Snyder & Sickmund, 1999). The propor-
tion by juveniles, a significant number of young people remain tion of young people who self-report having committed serious
both perpetrators and victims of interpersonal violence (Dahlberg, acts of violence has also held steady since peaking in the early
1998; Mercy & Potter, 1996; Sickmund, Snyder, & Poe-Yamagata, 1990s (Snyder, 2000).
1997; Snyder & Sickmund, 1999). For example, though the overall Violence occurs at home, in neighborhoods, and at school.
homicide rate in the United States has declined, rates for homicide Many recent studies illustrate the impact that exposure to violence
and nonfatal injuries among children and adolescents remain at and victimization from violence have on mental health and behav-

Daniel J. Flannery and Albert K. Liau, Institute for the Study and Injury Prevention and Control, Centers for Disease Control and Prevention,
Prevention of Violence, Kent State University; Alexander T. Vazsonyi, Atlanta, Georgia.
Department of Human Development and Family Studies, Auburn Univer- We gratefully acknowledge the contributions of Laura Williams, Kelly
sity; Shenyang Guo, School of Social Work, University of North Carolina Wester, Laurie Biebelhausen, and Lara Belliston to data management and
at Chapel Hill; Kenneth E. Powell, Centers for Disease Control and analysis. We also appreciate the comments of Thomas Simon on an earlier
Prevention, Atlanta, Georgia; Henry Atha, Pima County Community Ser- version of the manuscript. We are grateful to the students, staff, and parents
in the Sunnyside and Tucson unified school districts for their ongoing
vices Department, Tucson, Arizona; Wendy Vesterdal, Department of
support and participation.
Family and Consumer Resources, University of Arizona; Dennis Embry,
PeaceBuilders is a registered trademark of Heartsprings, Inc. The use of
PAXIS Institute, Tucson, Arizona.
trade names is for identification only and does not constitute endorsement
Albert K. Liau is now at the Psychological Studies Group, National by the U.S. Public Health Service or the U.S. Department of Health and
Institute of Education, Singapore. Kenneth E. Powell is now at the Georgia Human Services.
Department of Human Resources, Division of Public Health, Atlanta, Correspondence concerning this article should be addressed to Daniel J.
Georgia. Flannery, Institute for the Study and Prevention of Violence, Kent State
This project was supported in part by Cooperative Agreements U81- University, 230 Auditorium Building, Kent, Ohio 44242. E-mail:
CCU010038 – 03 and U81-CCU513508 – 01 from the National Center for dflanne1@kent.edu

292

SPECIAL ISSUE: VIOLENCE PREVENTION 293

ior, including an increased risk for engaging in violent behavior Farrington, 1998; Tolan & Gorman-Smith, 1998; Tremblay et al.,
(Elliott, Hamburg, & Williams, 1998; Flannery, 1997; Singer, 1992, 1995; Walker, Colvin, & Ramsey, 1995) and for becoming
Anglin, Song, & Lunghofer, 1995; Singer et al., 1999). Although antisocial adults (Eron & Huesmann, 1990).
the risk of homicide victimization at school remains low (Kachur Promising studies exist showing that the developmental trajec-
et al., 1996), levels of exposure to violence and victimization from tory of youth violence may be altered (CPPRG, 1999; Dahlberg,
violence at school remain high, particularly for elementary and 1998; Englander-Golden, Jackson, Crane, Schwarzkopf, & Lyle,
middle school children (Kaufman et al., 2000; Singer et al., 1999). 1989; Hawkins, 1995; Howard, Flora, & Griffin, 1999; Reid,
While recent data suggest a decline in the number of students Eddy, Fetrow, & Stoolmiller, 1999; Stoolmiller et al., 2000; Trem-
carrying weapons to school (7% of high school students were blay et al., 1991). Several studies have now demonstrated that
found to have done so within the previous 30 days; Kann et al., aggressive behavior can be reduced by altering the social environ-
2000), the use of firearms and other weapons has heightened the ments at school (Farrell & Meyer, 1997; Gottfredson, 1997;
lethality of violence among young people (Rushforth & Flannery, Greenberg, Kusche, Cook, & Quamma, 1995; Grossman et
1999) and has significantly increased the likelihood that specific al.,1997; Reid et al., 1999; Stoolmiller et al., 2000), particularly by
conflicts will escalate into lethal exchanges (Fagan & Wilkinson, emphasizing rewards and praise for prosocial behavior (CPPRG,
1997). In fact, despite recent declines in gun use and lethal forms 1999; Walker et al., 1995) and improving social competence
of violence, the proportion of young people involved in nonfatal (Hawkins, Catalano, Kosterman, Abbott, & Hill, 1999; O’Donnell,
violence has not declined (Snyder, 2000). Arrest rates for aggra- Hawkins, Catalano, Abbott, & Day, 1995) while reducing cues that
vated assaults remain almost 70% higher than they were in 1983, might increase hostility (Lochman & Dodge, 1994).
and this is the offense most frequently captured in self-reports The Good Behavior Game (GBG) is one type of school-based
of violence (U.S. Department of Health and Human Services prevention program that has established clear evidence of reduced
[USDHHS], 2001). aggressive behavior and other forms of child and adolescent prob-
The data are clear. Violence among young people remains a lem behavior such as tobacco use and poor academic achievement
significant public health problem (USDHHS, 2001). Although (Kellam & Anthony, 1998; Kellam, Ling, Merisca, Hendricks, &
many school and community-based violence prevention programs Ialongo, 1998; Salend, Reynolds, & Coyle, 1989). The GBG uses
exist, relatively few have been rigorously evaluated (Sherman et
classroom behavior management as the primary means of reducing
al., 1997; Thornton, Craft, Dahlberg, Lynch, & Baer, 2000). If
aggression and problem behavior. Student teams are rewarded by
psychologists are to inform public policy and facilitate risk pre-
teachers if no member of a team exhibits undesirable behaviors
vention for young people, it is imperative that we identify, through
while engaged in game sessions. Teachers begin by rewarding
applied evaluation studies, programs that effectively prevent youth
teams with tangible reinforcers and then gradually move to less
violent behavior and its associated precursors (i.e., aggression) and
tangible rewards.
rigorously evaluate the behavioral outcomes associated with these
The Linking the Interests of Families and Teachers (LIFT)
interventions (Powell & Hawkins, 1996; Satcher, Powell, Mercy,
program has also used the GBG as a core element of a 10-week
& Rosenberg, 1996; USDHHS, 2001). In the present study, we
universal preventive intervention strategy to reduce aggression and
examined the potential impact of a universal elementary-school-
increase social competence (Reid et al., 1999; Stoolmiller et al.,
based violence prevention program on student aggression and
social competence. 2000). The overall intervention consisted of parent training, the
GBG program, and systematic communication between teachers
and parents. The intervention had immediate and significant ef-
Preventive Interventions fects on physical aggression among students on the playground as
The need to provide early prevention is illustrated by the mul- well as some impact on increased child social competence. Exam-
titude of studies that show that violent behavior occurs along a ination of the LIFT program outcomes has also shown strong
developmental continuum of behavioral severity (e.g., Flannery & differential effects of treatment, with children highest on aggres-
Huff, 1999; Flannery & Williams, 1999; Tolan, Guerra, & Ken- sion at baseline benefiting the most from the intervention (Reid et
dall, 1995; Tremblay, et al., 1992). The precursors to more serious al., 1999; Stoolmiller et al., 2000). These short-term and differen-
violence perpetration in adolescence (e.g., homicide, assault) are tial effects on aggression are important to consider given the
young children’s aggressive behaviors such as hitting, kicking, and pressure to demonstrate significant behavior change with relatively
verbal insults and threats (Conduct Problems Prevention Research brief school-based interventions.
Group [CPPRG], 1999; Dahlberg, 1998; Huesmann et al., 1996; Another program of research has been conducted by the Con-
Huesmann & Moise, 1999; Singer & Flannery, 2000; Stoolmiller, duct Problems Prevention Research Group (CPPRG, 1999). The
Eddy, & Reid, 2000; Tremblay, Pagani-Kurtz, Masse, Vitaro, & CPPRG has implemented and evaluated the Fast Track prevention
Pihl, 1995). These are the triggers that can escalate interpersonal trial for conduct problem behavior for elementary school children
conflict into violence and are the behaviors that need to be targeted at high risk for long-term antisocial behavior. Fast Track is a
in preventive interventions in elementary schools. Young people developmentally based, long-term multicomponent and multisite
without the skills and competencies to resolve conflicts or solve intervention that has been evaluated using a randomized design
problems are at increased risk for violence victimization and with a nonintervention control group. After 1 year of intervention
perpetration (Lochman & Dodge, 1994). Longitudinal research has (from kindergarten to Grade 1), the group found moderate positive
consistently demonstrated that aggressive, peer-rejected children effects on children’s social competence and conduct problems,
in first grade are at increased risk for engaging in delinquent, including child aggressive behavior, for children in the interven-
violent behavior in adolescence (Hawkins et al., 2000; Loeber & tion schools compared with children in the control group.

294 FLANNERY ET AL.

Several other programs of research have demonstrated child All children and staff in a school learn five simple rules via a
behavior change in the areas of improved social competence or common language, which makes the intervention easy to learn and
reductions in aggressive behavior in the classroom or on the maintain: (a) praise people, (b) avoid put-downs, (c) seek wise
playground (e.g., Grossman et al., 1997; Hawkins et al., 1999; people as advisers and friends, (d) notice and correct hurts we
Tremblay et al., 1995; also see Thornton et al., 2000; USDHHS, cause, and (e) right wrongs. To help students learn these principles,
2001). In the current study, we sought to expand on these studies PeaceBuilders includes (a) daily rituals related to its language and
by examining the effects of an early elementary-school-based principles that are meant to foster a sense of belonging; (b) cues
universal preventive intervention program called PeaceBuilders. and symbols that can be applied to diverse community settings; (c)
This program attempts to alter individual child behavior—in par- specific prompts to “transfer” across people, behaviors, and time;
ticular, to reduce aggressive behavior and increase social compe- and (d) new materials or strategies introduced for times and
tence— by changing the culture or climate of an entire school. circumstances when positive behavior might otherwise decay (Em-
There is some evidence that PeaceBuilders affects the incidence of bry, 1980; Embry et al., 1996; Stokes & Baer, 1977).
assault-related and violent injury. Specifically, Krug and col- For example, staff and students are encouraged to use “praise
leagues (Krug, Brener, Dahlberg, Ryan, & Powell, 1997) found notes” to pay attention to and reinforce positive, prosocial behavior
that the frequency of injuries due to fighting for children in Grades in the classroom, at school, and at home. “Peace feet” might be
K–5 whose schools were randomized to PeaceBuilders did not placed by the drinking fountains to encourage children not to cut
increase over a 1-year period, although the incidence of injuries in line while waiting their turn, and students are sometimes sent to
due to fighting for children in control schools increased 56% over the principal for kind acts or good deeds rather than just for
the same period. Although these are meaningful archival data, we discipline problems (principal “preferrals”). PeaceBuilder rules
report here on teacher and child self-reports of social competence and principles are prominently displayed throughout the school,
and aggression, which have high predictive value for long-term and students complete activities from a specially designed comic
prevention efforts (CPPRG, 1999; Tolan et al., 1995; Tremblay et book in which they are the designated hero (see Embry et al.,
al., 1995; Vazsonyi, Vesterdal, Flannery, & Belliston, 1999; 1996). Adults more actively monitor “hot spots” in school such as
Walker et al., 1995). School is a logical public health setting for lunchrooms and hallways in between activities, praising prosocial
changing the cognitive, social, and imitative characteristics of behavior. All of these strategies and activities are geared toward
children at risk for violence. For example, schools can be thought creating a positive climate and culture in the entire school, with an
of as large antecedent and reinforcement systems that can increase emphasis on reinforcement of positive behavior rather than simply
or decrease antisocial and prosocial behavior (Mayer & Sulzer- the reduction of negative behavior.
Azaroff, 1990). We still lack consistent evidence of whether a The training of teachers in the implementation of the present
relatively low-cost, widely implemented universal preventive in- intervention had several phases, including a preintervention orien-
tervention approach in the early elementary grades will lead to tation for all faculty and staff of the schools, a half-day training
significant and sustainable behavior change. workshop on the basic PeaceBuilders model, and extensive site
coaching (on average, 2 hr per week) in the first 3 to 4 months of
The PeaceBuilders Program the intervention and then on an as-needed basis. All training and
coaching were conducted by the model developer (Embry et al.,
PeaceBuilders is a universal school-wide violence prevention 1996) as a means of facilitating internal validity. Each participat-
program for elementary schools (Grades K–5) implemented by all ing school also received specific in-service sessions on important
staff and students in a school (Embry, Flannery, Vazsonyi, Powell, issues identified by staff (e.g., implementing activities with special
& Atha, 1996). PeaceBuilders focuses on individual behavior needs children), periodic group forums to discuss successes and
change in proximal interpersonal and social settings (Tolan & challenges to implementation, and occasional 1-day institutes that
Guerra, 1994). The program incorporates an ongoing, long-term focused on applying and creating new materials and interventions.
strategy to alter the climate and culture of the entire school (Embry Attendance was voluntary at the institutes and forums. Additional
& Flannery, 1999; Embry et al., 1996; Flannery, 1997). The description of program materials and training is available else-
intervention is purposely woven into the school’s everyday routine where (e.g., Embry et al., 1996).
rather than presented as a time- or subject-limited curriculum.
Thus, PeaceBuilders is not offered as a set number of sessions or Hypotheses
hours per week but includes activities that can be implemented on
a daily basis in any classroom by any teacher or staff person. Our hypothesis was that youth aggressive behavior would be
Specifically, PeaceBuilders attempts to change characteristics of reduced by initiating prevention early in childhood and by increas-
the setting (antecedents) that trigger aggressive, hostile behavior, ing children’s resilience and social competence. A dual focus on
and it increases the daily frequency and salience of both live and reducing aggression and increasing social skills and competencies
symbolic prosocial models. If there are more prosocial cues and is important because the prognosis for children with a combination
models in a school and these behaviors are consistently reinforced of low social competence, aggressiveness, and poor emotional and
and rewarded, then over time, child social competence will in- cognitive preparation is poor (CPPRG, 1999; Kellam, Mayer,
crease and the frequency and intensity of aggressive behaviors will Rebok, & Hawkins, 1998; Tolan et al., 1995; Weissberg & Bell,
decline. PeaceBuilders specifically rewards prosocial behaviors 1997). We also examined the differential effectiveness of the
and provides strategies to avoid the differential or accidental intervention given evidence that treatment outcome effects may
reinforcement of negative behaviors and conflict that sometimes vary depending on a child’s initial behavior status prior to partic-
happens with conflict mediation programs (Webster, 1993). ipating in an intervention (Reid et al., 1999; Stoolmiller et al.,


2000). We examined both short-term change in aggression and Time 2 in the spring of Year 1. All participating schools remained in the
competence (compared with controls) over the 1st year of inter- study through the first 2 intervention years.
vention and longer-term change in Year 2, when all schools
received intervention. Specifically, in the 1st year, we expected Design and Procedure
that children in the intervention-school group, compared with
those in the control-school group, would report greater improve- Prior to baseline data collection, the eight project schools were matched
into four pairs primarily on the basis of geographic proximity, but we also
ments in social competence and greater reductions in aggressive
considered the percentage of ethnic students, the percentage of students
behavior. By the end of the 2 school years, when both groups were
eligible for free or reduced-price lunch, and the percentage of students in
receiving the intervention, we expected that, relative to baseline English as a Second Language (ESL) classrooms (see Table 1). School 2A
levels, students in both conditions would exhibit significant in- contained fewer Hispanic and more Native American students than its
creases in competence and prosocial behavior and decreases in comparison School 2B, but these schools were paired because of their close
aggressive behavior. geographic proximity. Four schools were then randomly assigned as
PeaceBuilders immediate intervention (PBI) schools and began the pro-
Method gram in the fall of 1994 immediately following baseline data collection.
The remaining schools began the PeaceBuilders program in 1995 after 1
The study protocol was approved by the Institutional Review Board for year of baseline data collection and are hereafter referred to as PeaceBuild-
Human Subjects at the University of Arizona in Tucson and by the ers delayed (PBD) schools (see Figure 1). PBD schools received compen-
respective schools’ research review committees. Parents were notified of sation in Year 1 ($1,000) as an incentive for them not to engage in any
the project through letters mailed to their homes and by school-distributed PeaceBuilders program-related activities.
newsletters. Parents were given the opportunity to withdraw their child We randomized at the school level because all students and staff in a
from any data collection. Students were also informed that their participa- school were exposed to and participated in the intervention. Students in the
tion was voluntary and were provided an opportunity for alternative class- four PBI schools were exposed to PeaceBuilders for a total of 2 school
room activities if they chose not to take part. If a student was engaged in years, and PBD schools participated in the intervention for 1 school year
another activity (e.g., band class), we returned to attempt to gather infor- between the fall and spring semesters of Year 2. Owing to limited re-
mation at a later date. At the time of survey administration, students were sources, we did not collect any child self-report data from new kindergarten
asked to give oral assent, and questions were answered regarding their students in Year 2, we collected Grades 1 and 2 child self-reports only for
participation. All students received rewards such as stickers or pencils for students who had participated in Year 1, and we did not follow Year 1 fifth
completing the surveys and interviews. graders into sixth grade. Further, students new to PBI schools in Year 2 of
Eight elementary schools (Grades K–5) in Pima County, Arizona, were the intervention were not included in these analyses.
selected from two large school districts to participate on the basis of having Students in Grades 3–5 completed 100-item self-report surveys at each
high rates of juvenile arrests and histories of suspensions and expulsions. data collection point. Surveys were administered in classrooms of about 20
After we met with school administrators to discuss the purpose and scope students with at least two research assistants present to read the entire
of the study, all schools that were initially contacted agreed to participate. survey aloud and to answer questions. This procedure resulted in few
Schools were located in all areas of town, including some in the central city surveys with missing or incomplete data. Surveys were pilot tested in two
and others on the outskirts of town. One of the eight schools consisted of elementary schools prior to data collection to assess the appropriateness of
a pair of schools in the same neighborhood, a school for Grades K–2 and the items for young children. All child survey items were answered with
a school for Grades 3–5 (approximately 1 block apart), and was treated as the anchors no, a little, or a lot.
a single school for pairing, intervention, analysis, and discussion (School For students in Grades K–2, self-report data were collected through
2A). All of the other schools were self-contained Grades K–5 schools. One individual 20-item, face-to-face interviews. The 20 items were pilot tested
school that was randomly assigned to the delayed intervention condition with same-age children. Owing to time constraints (we were only able to
(School 1B) did not gather initial baseline data but joined the study at interview as many children as time permitted during a single class period),

Table 1
School-Level Demographic Characteristics (%) of Matched Pairs

Matched African Native Asian Free ESL
schools Caucasian American Hispanic American American luncha pairsb

1A (n 704) 63.3 9.7 22.7 0.6 3.7 55 5
1B (n 551) 62.5 14.6 18.5 1.9 2.5 58 8

2A (n 817) 11.6 0.2 33.5 54.6 0.3 94 6
2B (n 377) 29.4 5.2 62.2 1.7 1.4 60 29

3A (n 550) 8.8 2.8 74.4 13.4 0.6 60 29
3B (n 573) 4.8 0.8 91.8 2.5 0.3 94 68

4A (n 327) 28.0 2.8 65.9 2.1 1.3 89 28
4B (n 780) 36.0 3.5 58.5 1.0 1.0 73 21

Note. “A” schools are those randomly assigned to the PeaceBuilders immediate (PBI) intervention, which
occurred immediately after baseline data collection. “B” schools were assigned to the PeaceBuilders delayed
(PBD) condition. a Percentage eligible for federally funded free or reduced-price lunch programs. b Students
for whom English was their second language.

296 FLANNERY ET AL.

Figure 1. Overview of project design, data collection, and intervention schedule.

we randomly preselected 50% of students in each kindergarten, first-grade, 1,101) of students reported that “Mom” took care of them the most, 15%
and second-grade class to be interviewed. Individual interviews, which reported “Dad,” 7% reported some other relative, and 2% each reported a
took about 5– 8 min to complete, were conducted at a table outside of the stepparent or some other adult. According to parent reports at Time 2 (n
child’s classroom in a quiet area. In Grades K–2 in the participating 809), 63% of children lived in homes with both parents present, 16% were
schools, the classes averaged about 20 students per classroom. By ran- from mother-only homes, and 12% lived with “one parent and other
domly preselecting half, we were attempting to target about 10 students per adults.” Parent reports of household incomes, although based on a sub-
class. Although there were no refusals of children in Grades K–2 to sample of our families, were evenly distributed among the lower range of
participate, on average we were able to complete 8 interviews per class, for socioeconomic groups: 22% reported an annual household income of
an effective participation rate of 80%. Reasons for not interviewing all 10 $7,000 or less; 19%, an income between $7,000 and $15,000; 24%, an
children included the following: Students were absent on the day of data income between $15,000 and $25,000; 23%, an income between $25,000
collection; students were engaged in an alternative school activity during and $40,000; and 12%, an income greater than $40,000 per year. The
the time interviews were conducted (e.g., band class), or we ran out of time. majority of our parents had completed the equivalent of high school or less:
Limited time and resources precluded our being able to interview children 15% completed less than ninth grade; 12% completed less than high
at a later date. Self-report interviews in Year 2 were conducted only for school; 28% completed high school; 38% completed some college; and 7%
available students who were interviewed in Year 1. Teachers continued to completed 4 or more years of college. Compared with 1990 U.S. Census
report via surveys on all kindergarten, first-, and second-grade children in data, our sample was similar to the population of the metropolitan area
their classrooms. (Pima County, AZ) on family composition, household income, and parent
At the time of each data collection, teachers of children in Grades K– 5 level of education. The only exception was for child ethnicity. In general,
completed a 45-item instrument for each student in their classes. Teachers our sample comprised higher percentages of minority children (and thus
provided written consent prior to participation. For Grades 3–5, both fewer Caucasians) than were in the greater metropolitan area from which
students and teachers answered questions on bubble scan sheets that the sample was drawn.
contained preassigned identification codes for data-tracking purposes. No Student and teacher sample sizes are reported in Figure 2. Student
names appeared on student data collection instruments. The preassigned ID response rates ranged from 86% to 93%, and teacher response rates from
code allowed us to distribute numbered surveys to specific students on the 75% to 86%. Fewer than 1% of parents chose to withdraw their child from
day of data collection as well as to link student and teacher data over time. any of the data collections. Similarly, fewer than 1% of children available
All student and teacher surveys were available in both English and Spanish. at each data collection time refused to complete a survey or interview,
Schools received compensation for their general funds depending on the usually citing disinterest.
percentage of teachers who completed surveys (e.g., $300 for 90% teacher
participation). Variables and Instrumentation
Sample Demographic variables. Demographic information gathered from stu-
dents included age, gender, and grade in school. Teachers reported on
On average, students across the 2 years examined were mostly Hispanic children’s ethnicity by categorizing them into one of six groups: Hispanic,
(51%), followed by Caucasians (28%), Native Americans (13%), African Caucasian, Native American, African American, Asian American, and
Americans (6%), and Asian Americans (1.5%). Seventy-one percent (n other.


Figure 2. Student and teacher sample sizes at each data collection point. The unit of randomization was the
school. aOf the children selected to be sampled, only 50% of the students in Grades K–2 were targeted to
participate in the child self-report portion of the study.

Aggressive behavior. Teachers reported on child aggressive behavior 1991; Grossman et al., 1997). The 25-item Aggressive Behavior subscale
using items adapted from the Aggressive Behavior subscale of Achen- asks teachers to rate child behavior on a 3-point scale in which 0 not
bach’s (1991) Teacher Report Form (TRF). The TRF has been used true, 1 somewhat or sometimes true, and 2 very true or often true. The
extensively as both a clinical screening instrument and in large survey items demonstrated high internal reliability ( .95 at baseline) in our
research to assess child externalizing behavior problems (Achenbach, sample.

Vazsonyi
Youth Violence EVALUATION OF
ARTICLEet al. / and Juvenile
10.1177/1541204003262224 Justice A VIOLENCE PREVENTION PROGRAM

EVALUATION OF A
SCHOOL-BASED, UNIVERSAL
VIOLENCE PREVENTION PROGRAM:
Low-, Medium-, and High-Risk Children

Alexander T. Vazsonyi
Lara M. Belliston
Auburn University
Daniel J. Flannery
Kent State University

The current investigation examined the differential effectiveness of PeaceBuilders, a
large-scale, universal violence prevention program, on male and female youth identi-
fied as low, medium, or high risk for future violence. It included eight urban schools ran-
domly assigned to intensive intervention and wait-list control conditions. The current
sample included N = 2,380 predominantly minority children in kindergarten through
fifth grade. Results indicated differential effectiveness of the intervention, by level of
risk; high-risk children reported more decreases in aggression and more increases in
social competence in comparison to children at medium and low levels of risk. Findings
add to a growing number of promising science-based prevention efforts that seek to
reduce aggression and increase social competence; they provide encouraging evidence
that relatively low-cost, schoolwide efforts have the potential to save society millions in
victim, adjudication, and incarceration costs.

Keywords: aggression; social competence; violence prevention; ethnicity

Young people are the primary perpetrators, victims, and often witnesses of interper-
sonal violence in our society (Snyder & Sickmund, 1999). Children who live in a climate of
violence learn to suppress empathy and learn that violence is an acceptable means to achiev-
ing their goals (Beland, 1996). This growing problem is evident in national crime statistics.

Authors’ Note: A previous version of this article was presented at the 9th Biennial Meetings of the Society for
Research on Adolescence in New Orleans (April 2002). This project was supported in part by cooperative agree-
ments from the National Center for Injury Prevention and Control, Centers for Disease Control and Prevention
(#U81-CCU010038-03 and #U81-CCU513508-01), Atlanta, GA. We would like to thank Dennis Embry as well
as students, staff, and parents in the Sunnyside and Tucson Unified School Districts for their participation.
PeaceBuilders is a registered trademark of Heartsprings, Inc. The use of trade names is for identification only and
does not constitute endorsement by the Public Health Service or the U.S. Department of Health and Human Ser-
vices. Please address correspondence related to this article to Alexander T. Vazsonyi, Ph.D., Dept. of Human
Development and Family Studies, 284 Spidle Hall, Auburn, AL 36849; e-mail: vazsonyi@auburn.edu.
Youth Violence and Juvenile Justice, Vol. 2 No. 2, April 2004 185-206
DOI: 10.1177/1541204003262224
© 2004 Sage Publications

185

186 Youth Violence and Juvenile Justice

Even though recent publications of the National Crime and Victimization Survey (NCVS)
and Uniform Crime Reports (UCR) note continued decreases in violence over the past sev-
eral years (U.S. Department of Justice [USDOJ], 2001a, 2001b), juvenile violence remains
high. Although firearm-related homicides have decreased, the youth homicide rate has
remained fairly stable; moreover, the overall youth violence index, assault with injury, and
robbery with a weapon have increased (U.S. Department of Health and Human Services
[USDHHS], 2001). In addition, a cross-national comparison shows that the rate of adoles-
cent homicides involving a firearm is over 15 times higher in the United States than in 12
European countries combined (Centers for Disease Control and Prevention [CDC], 1997).
Prevention and intervention efforts designed to ameliorate violence have identified a
number of individual, family, school, peer, and community risk factors that contribute to
delinquency and future violence (Andrews & Trawick-Smith, 1996; Consortium on the
School-Based Promotion of Social Competence, 1994). Although many of these factors can
help identify individuals at risk for problem behaviors, good prevention efforts need to tar-
get risk factors most amenable to change, such as skills training, behavior monitoring and
reinforcement, behavioral techniques for classroom management, and building school
capacity (USDHHS, 2001). A number of individual-level risk factors can be targeted by
violence prevention programs. Such factors include general offenses, substance use,
aggression, problem behaviors, and antisocial attitudes (Gottfredson, 2001; USDHHS,
2001). Several of these risk factors are highly confounded with rates of deviance; however,
the most salient behavioral predictor of later violence and delinquency is early aggression
between ages 8 and 10 years (Farrington, 1987; Gottfredson, 2001; Hawkins et al., 1998;
Lipsey & Derzon, 1998; Loeber & Dishion, 1983; O’Donnell, Hawkins, & Abbott, 1995;
Viemerö, 1996; USDHHS, 2001). Furthermore, aggression in the school context is highly
problematic during grade school as it violates peer group and social norms (Bierman &
Montminy, 1993; Coie & Dodge, 1998). Cross-sectional research has demonstrated that
childhood aggression can foretell official delinquency status (Vazsonyi, Vesterdal,
Flannery, & Belliston, 1999). Longitudinal investigations have also demonstrated that
aggressive behavior is relatively stable over time and part of a general pattern of antisocial
behavior that is associated with later self-reported violence, arrests, and convictions for vio-
lent offenses (Farrington, 1987; Lipsey & Derzon, 1998; Loeber, Farrington, Stouthamer-
Loeber, Moffitt, & Caspi, 1998; Viemerö, 1996).
Most violence and delinquency prevention research has focused on reducing aggres-
sion; however, researchers have also emphasized protective factors that may interact with
risk factors to buffer or reduce risk of future violence (Bierman, Miller, & Stabb, 1987; Coie
& Koeppl, 1990; USDHHS, 2001). Individual-level factors that protect against delin-
quency include a positive social orientation and an intolerant attitude toward interpersonal
violence and deviance (USDHHS, 2001). Recent prevention efforts have targeted behav-
ioral measures of social competence and prosocial skills (e.g., Blechman, 1996; O’Donnell
et al., 1995). Children who lack these skills are more likely to rely on their negative pat-
terns of interaction and demonstrate more negative behavioral outcomes (Ollendick, Weist,
Borden, & Greene, 1992; Quinn, Mathur, & Rutherford, 1995; Walker & McConnell,
1988). However, few rigorous evaluation studies have been completed examining risk and
protective factors of juvenile violence and deviance.
A next important step for researchers is to identify how risk and protective factors
work together to influence problem behaviors. Kupersmidt, Coie, and Dodge (1990) found
that aggressiveness and social competence predicted delinquency in elementary school.
Similarly, Hämäläinen and Pulkkinen (1995) found that rates of recidivism were greater for

Vazsonyi et al. / EVALUATION OF A VIOLENCE PREVENTION PROGRAM 187

criminals who had been more aggressive and less prosocial when they were young. Con-
versely, in a review of intervention programs, Coie and Koeppl (1990) observed that a num-
ber of programs targeting aggressive, disruptive, and rejected children focused their pre-
vention efforts on increasing prosocial behaviors and paid insufficient attention to reducing
aggressive behaviors (e.g., Frey, Hirschstein, & Guzzo, 2000; Gottfredson, Gottfredson, &
Skroban, 1998; Prinz, Blechman, & Dumas, 1994). Both behaviors must be changed to
achieve intervention effectiveness (Blechman, 1996; Coie & Koeppl, 1990; Gresham &
Elliott, 1987; USDHHS, 2001; Wasserman & Miller, 1998). Therefore, rigorous evaluation
studies of programmatic efforts should focus on risk and protective factors, and they should
evaluate how both of these behaviors change following an intervention.

School-Based Violence Prevention

Recent violence prevention efforts have shifted to large-scale, universal program-
matic efforts (Powell et al., 1996). Although prevention efforts have occurred in multiple
contexts, school-based interventions have several advantages (Catalano, Arthur, Hawkins,
Berglund, & Olson, 1998; Gottfredson, 2001; USDHHS, 2001). For example, schools are
an optimal setting for preventions and interventions; children spend a great deal of time at
school with teachers and peers, and large groups of at-risk children can be easily targeted
(Beland, 1996; Blechman, 1996). Effective strategies for universal school implementation
include behavioral monitoring and reinforcement, classroom management, and skills train-
ing; students receive direction from their primary teacher and support from other school
staff members. This approach recognizes that behavior change takes time; it also recognizes
that the total school atmosphere needs to change as reinforcements are implemented across
school experiences (e.g., Farrell, Meyer, Kung, & Sullivan, 2001; Gottfredson, 2001;
USDHHS, 2001).
Several large-scale, school-based violence prevention programs targeting elementary
school students have documented promising findings of program effectiveness (cf. the Stu-
dents for Peace Project, Kelder et al., 1996; Orpinas et al., 2000). For example, the Resolv-
ing Conflicts Creatively Program (RCCP) (Aber, Jones, Brown, Chaudry, & Samples,
1998) found that the program did not reverse negative or positive behavior patterns but sig-
nificantly slowed the trajectories for increasing aggressiveness and decreasing social com-
petence, particularly for students who were exposed to most of the programmatic compo-
nents. Similarly, findings from the Fast Track prevention trial by the Conduct Problems
Prevention Research Group (CPPRG) indicated that the program has decreased rates of
conduct problems in children identified as being at high risk for behavior problems in kin-
dergarten (baseline; 27% children with conduct problems in the intervention group vs. 37%
in the control group; CPPRG, 2002). In another effort evaluating the effects of Peacemak-
ers, Shapiro, Burgoon, Welker, and Clough (2002) found decreases in self-reported and
teacher-reported aggressive behaviors as well as decreases in the number of disciplinary
incidents and suspensions following program implementation. The study also indicated
stronger program effects for boys than for girls and for younger children than older ones.
Finally, teacher-reported data showed more consistent and stronger program effects than
student data, although self-reported student data corroborated findings based on teacher
reports.
Additional programs require some discussion. Again focusing on a high-risk sample
of children, the Metropolitan Area Child Study (MACS) (Eron, Huesmann, Spindler,


Guerra, & Henry, 2002; Guerra, Eron, Huesmann, Tolan, & Van Acker, 1997) provided
evidence of program effectiveness. Findings indicated that the program was most beneficial
when it was administered during the early school years and where it was supported by a 2-
year follow-up intervention. They also indicated that the intervention was equally effective
for boys and girls; in fact, although median levels of aggression increased over time in inter-
vention and control conditions, a significant number of children moved from clinical to
nonclinical status for externalizing behavior problems following the intervention.
The Responding in Peaceful and Positive Ways (RIPP) (Farrell & Meyer, 1997;
Farrell, Meyer, & White, 2001) program and evaluation study provided evidence of a
reduction in violent behaviors and less in-school suspensions following the intervention.
The reduction in violent behaviors was most evident in students who had high levels of vio-
lent behaviors at pretest, which indicated a differential programmatic effect.
Finally, two studies evaluating Linking the Interests of Families and Teachers (LIFT)
(Reid, Eddy, Fetrow, & Stoolmiller, 1999; Stoolmiller, Eddy, & Reid, 2000) found support
for reducing young children’s physical playground aggression and increasing teacher rat-
ings of peer-preferred behaviors. Differential effectiveness for reducing children’s aggres-
sion were found over time, namely, that children with the highest levels of aggression at
pretest showed more changes than children with lower pretest scores. To assess this differ-
ential effectiveness, Stoolmiller et al. (2000) measured the effect sizes at four levels of
aggression and found medium to high effect sizes for children with the highest levels of
aggression at pretest.
These findings are encouraging and are consistent with Durlak and Wells’ (1997)
meta-analysis that showed that programs targeting reducing negative behaviors and pro-
moting social competency show promise. These programs addressed a specific recommen-
dation by Durlak and Wells (1997) and Weissberg and Bell (1997) to evaluate program suc-
cess for at-risk populations. In particular, these studies started to address the differential
effectiveness of programs, how well the programs work for children at risk for future vio-
lence, rather than addressing main effects between intervention and control groups. Indeed,
Stoolmiller et al. (2000) identified differential effectiveness as a key issue for universal pro-
grams. However, researchers disagree on how to best determine risk for future delinquency.
Because of low base rates, only a small number of children become classified as officially
delinquent, approximately 5% to 6% of boys (Vazsonyi et al., 1999). RIPP, LIFT, and
PeaceBuilders have addressed differential effectiveness utilizing regression methodol-
ogy (Farrell, Meyer, & White, 2001; Flannery et al., 2003; Stoolmiller et al., 2000). In par-
ticular, previous research on PeaceBuilders found differential effectiveness for teacher-
reported aggression, self-reported aggression, and prosocial behavior.
Based on these studies, the purpose of this article is to test the differential effective-
ness hypothesis, namely, that programs have greater effects on children with high rates of
problem behaviors as opposed to children with very low rates. Rather than utilizing a
regression procedure, an alternative method for assessing differential effectiveness is to
assign children to risk levels. In addition, instead of classifying children at risk by official
delinquency status, risk determination should include more children by identifying vari-
ables that predict delinquency that are not confounded with measures of delinquency
(Loeber & Dishion, 1983). LeBlanc (1998) advocated using a “multiple-gating” procedure
developed first by Loeber and Dishion (1983) that uses several assessments or predictors as
screening gates. The first step is to apply the first predictor to the full sample, temporarily
classifying children into risk and nonrisk samples by the primary factor. Subsequently, in
the second step, children are maintained or dropped from the risk sample based on the sec-


ond predictor. The result is that a larger number of children are classified at risk for a partic-
ular outcome, which may be beneficial for determining how effective programs are for
children most at risk for future problems.
Thus, the current investigation examined the differential effectiveness of Peace-
Builders on children identified as low, medium, or high risk for future problems. Children
were classified by the multiple-gating procedure into three risk groups (low, medium, and
high risk) based on teachers’ assessments of aggression and social competence.

PeaceBuilders Violence Prevention Program

PeaceBuilders is a schoolwide, universal violence prevention program that is theoret-
ically based (Embry, Flannery, Vazsonyi, Powell, & Atha, 1996). The program attempts to
change antecedents that trigger aggressive behavior, reward prosocial behavior, and pro-
vide strategies to avoid reinforcing negative behavior. PeaceBuilders is organized around
five main principles: (a) PeaceBuilders praise people, (b) PeaceBuilders avoid put-downs,
(c) PeaceBuilders seek wise people, (d) PeaceBuilders notice hurts they have caused, and
(e) PeaceBuilders right wrongs. The intervention structure uses several behavior techniques
to promote change: symbolic and live models, role-plays and rehearsals, and group and
individual rewards.
The PeaceBuilders program was implemented in the school setting by teachers, prin-
cipals, and other support staff. members Teachers use a variety of materials to help teach
and encourage students to be PeaceBuilders: “I Help Build Peace” story/workbooks, medi-
ation essays, Praise Boards (written records of positive events), games (The Peace Scout
Game where anonymous scouts send secret notes), home notes, posters made by children,
PeaceCards and secret notes. Teachers received an hour-long preintervention orientation, 3
to 4 hr of training workshops, and 2 hr of site coaching per week that occurred during the
first 8 to 12 weeks of program implementation. Additional help sessions were offered when
schools had specific questions regarding PeaceBuilders (Embry et al., 1996).
The study design included nine project schools with children in kindergarten through
fifth grades. One Grade K-2 school and one Grade 3-5 school were combined to form a sin-
gle K-5 unit. These eight school units were then grouped into four matched pairs. Within the
pairs, schools were randomly assigned as intervention (Wave 1) or wait-list control (Wave
2) schools. Baseline data (Time 1) for all schools were conducted in the fall of 1994. Wave 1
schools received the intervention following baseline, in the fall of 1994, and Wave 2
schools received the intervention in the fall of 1995. Data collection occurred every fall and
spring for 2 years; Time 2 data collection occurred spring 1995, Time 3 data collection
occurred fall 1995, and Time 4 data collection occurred spring 1996.
A preliminary analysis of the effectiveness of PeaceBuilders compared children in an
initial treatment versus delayed treatment condition. Children in both conditions had similar
baseline levels of aggression and social competence. Hierarchical linear modeling deter-
mined that children who received the intervention in the initial treatment condition showed
significant increases in teacher-rated social competence and child self-reports of peace
building as compared to the delayed treatment condition, over a 2-year period. Similarly,
after 12 months, children in the delayed treatment condition showed significantly higher
rates of aggression than children in the continuous treatment condition (Flannery et al.,
2003). Whereas preliminary longitudinal analyses show some changes in children’s behav-


iors because of PeaceBuilders interventions, these results need to be more carefully exam-
ined, especially for children who vary in degree of risk for future violence at baseline.
Risk status established through a multiple-gating procedure has been used to examine
how level of risk subsequently predicts delinquency or externalizing behaviors in previous
studies (Lipsey & Derzon, 1998; Lochman & The Conduct Problems Prevention Research
Group, 1995; Loeber & Dishion, 1983; Patterson, Capaldi, & Bank, 1991); however, it has
not been utilized in evaluation research. One exception is the MACS project, where Guerra
et al. (1997) reported splitting children into two risk categories. The most recent reports
of MACS program effects (Eron et al., 2002) focused on program effects in the high-risk
group of children. Across violence prevention programs, program effectiveness has not
been evaluated at multiple levels of risk. Therefore, the current investigation cannot make
specific predictions about program effectiveness by level of risk based on previous work; at
the same time, we did expect the greatest amount of change in high-risk youth and the
smallest amount of change in low-risk youth.
More specifically, we expected differential program effects on behavioral outcome
measures by levels of risk. For example, we expected that children identified as high risk
would show the greatest program effects in teacher-reported aggression and social compe-
tence, and self-reported aggression and prosocial behavior. We also expected more modest
program effects for children identified at medium risk for future violence and delinquency,
and we expected very few program effects for children identified at low risk for future vio-
lence and delinquency. Finally, we expected that we would find similar programmatic
effects by levels of risk for boys and girls.

Method

Sample

The sample for the current study is based on the PeaceBuilders violence prevention
evaluation project conducted in the Tucson metropolitan area (Embry et al., 1996; Flannery
et al., 2003; Vazsonyi et al., 1999). The targeted region had experienced an increase in vio-
lent offenses from 1990 to 1993—increases in juvenile arrests for violent crimes and homi-
cides, vandalism, and weapons violations. Juvenile arrests for total, property, and violent
crimes continued to increase and peaked in 1995. Since 1995, juvenile arrests have been
decreasing. However, property crimes have decreased at a higher rate than violent crimes,
which are still high at similar levels as reported in 1990-1991 (Geospatial and Statistical
Data Center [Geostat], 2003). In addition to community-wide efforts to increase social and
cognitive competencies related to preventing violence, a comprehensive program, Peace-
Builders, was implemented within two city school districts.
Two school districts were chosen based on police crime maps; these maps identified
areas with high levels of violent crimes and high neighborhood stress (e.g., domestic vio-
lence, transition and mobility, poverty levels). Nine schools (one K-2 and 3-5 were com-
bined to form one school unit) were invited for participation based on these data (Embry
et al., 1996). Schools were matched into four pairs based on geographic proximity, student
ethnicity, percentage of students eligible for free or reduced lunch, and percentage of stu-
dents with English as their second language (Embry et al., 1996; see Table 1). It is important
to note that some of the matched school pairs differed on key variables (e.g., student ethnic-


TABLE 1
School Level Demographic Characteristics:
Percentage of Ethnicity and Socioeconomic Variables

School School School School School School School School
1A 1B 2A 2B 3A 3B 4A 4B
(n = 704) (n = 551) (n = 817) (n = 377) (n = 550) (n = 573) (n = 327) (n = 780)

African American 9.7 14.6 0.2 5.2 2.8 0.8 2.8 3.5
Asian/Pacific Islander 3.7 2.5 0.3 1.4 0.6 0.3 1.3 1.0
Hispanic 22.7 18.5 33.5 62.2 74.4 91.8 65.9 58.5
Native American 0.6 1.9 54.6 1.7 13.4 2.5 2.1 1.0
White 63.3 62.5 11.3 29.4 8.8 4.8 28.0 36.0
Free lunch 55 55 94 60 60 94 89 73
English as Second
Language 5 8 6 29 29 68 28 21

NOTE: Schools were matched in pairs based on geographic proximity, student ethnicity, percentage of students eli-
gible for free or reduced lunch, and percentage of students with English as their second language. Schools were ran-
domly assigned to the initial intervention condition (A schools) or the wait-list control condition (B schools).

ity). Schools were then randomly assigned to the initial intervention condition or the wait-
list control condition.
The sample included approximately 4,600 children from kindergarten through fifth
grade. Of the total sample, 50% were Hispanic, 29% White, 15% Native American, 5%
African American, and 1% Asian/Pacific Islander. Children were roughly evenly divided
by sex between the intervention and wait-list control conditions. Figure 1 graphically pres-
ents program intervention and data collection periods. One half of the schools received the
PeaceBuilders intervention in the fall of the first year (Wave 1), while the other one half
received the intervention during the following fall (Wave 2). For the current investigation,
we focused on children in third through fifth grades who had teacher ratings of aggression
and social competence and self-reports of aggression and prosocial behaviors at the base-
line (Time 1). Data from children in both treatment conditions (Wave 1 and 2) were aggre-
gated to permit a comparison of children’s pretest scores and posttest scores. Data at pretest
were Time 1 data for Wave 1 children and Time 3 data for Wave 2 children. Data at posttest
were Time 2 data for Wave 1 children and Time 4 data for Wave 2 children. ANOVAs com-
paring pretest scores by Wave showed only one significant difference for female prosocial
behavior; Wave 2 girls reported slightly higher prosocial behaviors at pretest than Wave 1
girls (F = 5.96, p < .05, d = .19).
The total number of students enrolled in project schools in kindergarten through
Grade 5 at initial data collection was N = 4,679. Some students were excluded from the sam-
ple because of incomplete data. Teacher data were available on children in kindergarten
through Grade 5. Complete teacher data (K-5) were collected from n = 2,380 children (M
age = 8.5 years), a response rate of 50.8% (see Table 2). Because of cognitive and language
ability, child self-report data were only available for Grades 3 through 5. Complete child
self-report data (3-5) were obtained for n = 1,170 children (M age = 9.8 years), a response
rate of 52.2% (see Table 2). The low response rates (as compared to Flannery et al., 2003)
are due largely to the construction of this sample. First, to classify the students into risk cat-
egories by teacher reports, the sample was limited to students with baseline teacher-
reported data. Second, to compare pretest and posttest data, Wave 1 children had to have


Figure 1. Overview of Project Design, Data Collection, and Intervention Schedule

Time 1 and Time 2 data, whereas Wave 2 children had to have Time 3 and Time 4 data.
Thus, students without child and/or teacher data at Time 1 and Time 3, for example, were
dropped from the sample. This selection process did not appear to vary by sex or race.
Approximately 50% of boys and girls were dropped, and race percentages ranged from
20% to 25% for all groups except Native Americans (41%). In addition to sample construc-
tion issues, subject attrition rates were related to relatively high residential mobility within
school districts.

Procedures

Data were collected by trained project staff members from teachers in Grades K
through 5, and children in Grades 3 through 5. During regular school hours, children com-
pleted in-class surveys administered by project staff members who read all questions aloud.
The survey took approximately 1 hr to complete, and students received small incentives for
their participation, such as stickers or pencils. Teachers completed surveys for each child in
their classroom. They received data collection packets at the time of the student survey data
collection. Each teacher received $20 for participation. In addition, the schools were eligi-
ble for schoolwide incentives based on the number of teacher surveys returned ($300 to
$500). During the initial phase of the project, data were collected at four points in time (two
fall and two spring, in consecutive school years) for schools assigned to two intervention
conditions. Schools in Wave 1 started the intervention immediately following baseline data
collection. Schools in Wave 2 began the intervention about 1 year later following Time 3
data collection.


TABLE 2
Sample Description and Response Rates

Low Medium High
Not Risk Risk Risk Total
Classified n n n n N %

Teacher report (K-5)
with baseline dataa (1,132) 1,147 1,099 1,301 3,547 4,679 75.8
Teacher report (K-5) with
pretest/posttest scoresb 840 746 794 2,380 4,679 50.8
Males 455 341 376 1,172 2,380 49.2
Females 385 405 418 1,208 2,380 50.8
Child self-report (3-5) with
pretest/posttest scoresb 443 332 395 1,170 2,243 52.2
Males 221 146 197 564 1,170 48.2
Females 222 186 198 606 1,170 51.8

Low Medium High
Not Risk Risk Risk
Classified % % % n Nd %
c
Ethnicity
African American 20.1 21.8 18.4 39.7 174 3,086 5.6
Asian/Pacific Islander 20.4 40.8 22.4 16.3 49 3,086 1.6
Hispanic 25.6 25.8 24.7 23.8 1,540 3,086 49.9
Native American 40.8 18.7 18.1 22.3 497 3,086 16.1
White 24.0 28.2 21.7 26.2 826 3,086 26.8

NOTE: a. Baseline data for all children were collected at Time 1.
b. Data at pretest were Time 1 data for Wave 1 children and Time 3 data for Wave 2 children. Data at posttest were
Time 2 data for Wave 1 children and Time 4 data for Wave 2 children.
c. Ethnicity statistics reported here were obtained from archival school records and are reported for students with
baseline data.
d. Ethnicity was available for 3,086 children out of the 4,679 children eligible with baseline data, 34.0% of the sam-
ple was missing data on ethnicity.

Measures

Social competence (teacher report). The 19-item short-form version of the Walker-
McConnell Scale of Social Competence (Walker & McConnell, 1995) measured social
skills and school adjustment as rated by teachers. The instrument has been used in long-
term follow-up studies and has predictive value, particularly for children with serious
behavior problems (Fifeld, 1987; Hops, 1987). The scale includes three subscales: School
Adjustment (e.g., “student attends to assigned tasks” and “produces work of acceptable
quality given his or her skills”); Peer Preferred Behavior (e.g., “invites peers to play” and
“shares laughter with peers”); and Teacher Preferred Behavior (e.g., “can accept not getting
his or her way” and “compromises with peers when a situation calls for it”). Teachers rated
each item on a 5-point scale from 1 = never to 5 = frequently (α = .95). The three subscales
were summed to produce an overall Social Competence score (Flannery et al., 2003;
Vazsonyi et al., 1999).


Aggressive behavior (TRF). Physical and nonphysical aggressive behavior was mea-
sured by the 25-item Achenbach’s Child Behavior Checklist Teacher Report Form
(Achenbach, 1991; Flannery et al., 2003; Vazsonyi et al., 1999). Teachers were asked to
recall children’s behavior over the past 2 months; examples include “The child argues a lot,”
“The child gets in many fights,” and “The child threatens people.” Responses were given on
a 3-point Likert-type scale, 0 = not true, 1 = somewhat true, or 2 = very true (α = .95).

Prosocial behavior (child report). This 16-item scale was developed by the research
team to measure how much children engaged in prosocial acts over the past 2 weeks
(Flannery et al., 2003; Vazsonyi et al., 1999). Children responded to questions such as “I
did things to help other kids,” “I smiled at others,” and “I apologized to a grown-up at
school.” Responses were given on a 3-point scale, 1 = no, 2 = a little, and 3 = a lot (α = .92).

Aggressive behavior (YSR). This scale consisted of nine items from the Delinquency
and Aggression subscales of the Child Behavior Checklist-Youth Self Report (Achenbach,
1991; Flannery et al., 2003; Vazsonyi et al., 1999). Questions asked about physical and
nonphysical aggression over the past 2 weeks, for example, “I teased other kids at school,”
“I hit someone,” and “I tried to get other students to fight.” Responses were given on a 3-
point Likert-type scale, 0 = no, 1 = a little, to 2 = a lot (α = .95).

Plan of Analysis

Initial descriptive statistics on teacher reports and child self-reports were computed
for all children with baseline (Time 1) data. These data were used to classify children into
three risk groups: low, medium, and high risk.
Analyses for the current study were computed using general linear modeling (GLM).
GLM covers a variety of linear models of analyses of variance and covariance, regression,
and repeated measures models (Howell, 1992); it also adjusts for unequal cell sizes and pro-
vides estimated marginal means (predicted estimates of the population marginal mean
based on regression; Searle, Speed, & Milliken, 1980). GLM repeated-measures proce-
dures account for variation in the pretest and posttest scores by computing a pooled value
for multivariate tests and subsequently determines change over time using estimated
marginal means (SPSS, 1999).
To maximize sample size, GLM analyses were conducted separately by sex and by
teacher and self-report data. Risk status was entered in the model as a between-subjects
variable with three levels of risk. Age and race were also included in the model as
covariates. Age was a continuous covariate, and race was a categorical covariate, namely,
White versus non-White. Because of the inclusion of covariates in the model, GLM analy-
ses were conducted in two steps (Winer, 1971). The first step ran the model with the
covariates and reported the between-subjects portion of the model. The second step ran the
model without the covariates and reported the within-subjects portion of the model. Subse-
quently, pairwise comparisons were conducted based on the estimated marginal means.
Because differences over time were hypothesized a priori, significant pairwise comparisons
were reported regardless of the significance of the omnibus F statistic (Girden, 1992;
Tabachnick & Fidell, 1989).
One assumption of ANCOVA is that the regression coefficients are equal (Howell,
1992). A significant covariate would violate that assumption, usually invalidating the use of


TABLE 3
Descriptive Statistics of Teacher Reports by Sex (N = 3,554)

Males Females
(n = 1,765) (n = 1,779)
M SD Median M SD Median

Social competence 3.49 .82 3.53 3.86 .76 3.89
Aggression (TRF) 1.45 .52 1.24 1.23 .38 1.04

NOTE: Social competence and aggression (TRF) are teacher-reports. TRF = Teacher Report Form of the
Achenbach Child Behavior Checklist (Achenbach, 1991).

ANCOVA for modeling data (Howell, 1992; G. Hudson, personal communication, March
8, 2002; SPSS, 2002). However, in the current study, the covariates included in the model
were not treatment effects but naturally occurring variations in the population. Therefore,
we were interested in the percentage of variance in the model explained by each of the
covariates (Howell, 1992). For GLM repeated measures, this was done in two procedures:
first by analyzing the slope and variance of the covariates for the pooled dependent variable
(consistent with GLM multivariate tests) and subsequently by examining whether the
covariate influenced change in the dependent variables over the course of the program by
using change scores (Howell, 1992; G. Hudson, personal communication, March 8, 2002;
SPSS, 2002; Tabachnick & Fidell, 1989). The second step of the analyses determined the
effect of the program over time for each of the three risk groups after controlling for the
effects of covariates. Pairwise comparisons of the estimated marginal means determined
programmatic effects on children’s behaviors at different levels of risk.

Results

At-Risk Status

Children’s at-risk status was determined by teacher-reported aggression and
social competence scores collected at baseline. Low, medium, and high risk was defined by
a multiple-gating procedure that utilized a median split of the two risk variables separately
by sex (see Table 3).
Boys and girls were classified as high risk if they had scores above the median in
aggression and below the median for social competence; in other words, they exhibited high
negative behaviors and few positive ones. Individuals were classified as low risk if they had
scores below the median score for aggression; these children also reported high social com-
petence scores. The medium-risk group was characterized by scores above the median in
aggression and social competence or by scores below the median in aggression and social
competence. In other words, these children exhibited a mixture of positive and negative
behaviors (see Figure 2 for the multiple-gating procedure; see Table 3 for the number of
children in each of the risk groups by sex; there were at least 125 children in each risk
category for boys and girls).


Figure 2. Multiple-Gating Procedure

Winer Model Step 1:
Inclusion of Covariates and Between-Subjects Variable

The first step of the Winer model includes both covariates, age and race, and the inde-
pendent variable, risk status. The results of the analyses are presented in Table 4. The first
three columns report the multivariate F statistics for each dependent variable. The next four
columns report the slope and percentage of variance explained by each of the covariates in
the pooled dependent variable. The last four columns report the slope and percentage of
variance explained by the covariates on the amount of change over time in each dependent
variable.
Significant effects for age were found for female social competence, male teacher-
rated aggression, male and female prosocial behaviors, and male self-reported aggression.
Significant effects for race were found only for female prosocial behavior as well as male
and female self-reported aggression. Significant effects for risk as a covariate were found
for male and female social competence, teacher-reported aggression, and self-reported
aggression. No significant effects for risk were found for prosocial behavior.
When analyzing the effects of the covariates on the pooled dependent variables, age
accounted for a large percentage of the variance in prosocial behavior (8.9% girls, 12.8%
boys). For both of these variables the slopes were negative; as age increased, prosocial
behavior decreased. Age also accounted for 4.4% of the variance in male teacher-rated
aggression and 3.2% of the variance in female social competence. Race accounted for very
little variance in the dependent variables; the highest percentage of variance attributed to
race was for male self-reported aggression (1.4%). Even though age and race accounted for
some proportion of the variance in the pooled dependent variables, additional analyses
needed to determine the percentage of variance these covariates explained in pre/post
change scores. In these scores, age accounted for a small proportion of the variance in male
(2.5%) and female (1.5%) changes of social competence and male self-reported aggression
(1.5%). Race accounted for very little variability in change scores, namely, 1.7% of the

TABLE 4
2
Winer Model (Step 1): Between-Subjects F Values, and Covariate Analyses Slope, and R Values for General Linear Modeling (GLM) model

Pooled DV Change Scores
Slope % Slope % Slope % Slope %
Age Race Risk of Age Variance of Race Variance of Age Variance of Race Variance
d d
Fd F F β Age β Race β Age β Race

Males
a
Social competence 2.45 1.35 381.40* –.001 1.82 –.004 0.07 –.063 2.54 .014 0.00
a,c
Aggression (TRF) 6.21* 1.09 342.29* –.001 4.43 –.002 0.04 .001 0.11 .003 1.69
Prosocial behaviorb 108.11* 0.24 1.33 –.19 12.78 –.002 0.04 –.023 0.19 .037 0.11
Aggression (YSR)b,c 6.95* 7.64* 18.72* –.005 1.36 –.106 1.42 .056 1.53 .047 0.24
Females
Social competencea 6.39* 1.67 318.00* –.002 3.23 –.004 0.05 –.044 1.45 .001 0.00
Aggression (TRF)a,c 0.29 0.82 208.51* .000 0.00 .002 0.04 .016 0.78 .047 0.54
Prosocial behaviorb 65.60* 4.40* 0.95 –.152 8.91 –.008 0.57 –.004 0.00 –.007 0.00
Aggression (YSR)b,c 2.04 5.89* 11.83* –.002 0.55 –.005 0.83 –.003 0.00 .037 0.32

a. Teacher reports of social competence and aggression are listed first, n ranges from 323 to 411.
b. Child self-reports of prosocial behavior and aggression, n ranges from 125 to 194.
c. TRF = Teacher Report Form of Achenbach’s Child Behavior Checklist; YSR = Youth Self-Report Form of Achenbach’s Child Behavior Checklist (Achenbach, 1991).
d. Multivariate F statistic is significant at *p < .05.

197


variance in male teacher-reported aggression changes. The second step of the analyses
demonstrated significant changes over time in each of the three levels of risk. Therefore, a
series of analyses was completed in the second step of the Winer model; results are reported
in Table 5.

Winer Model Step 2:
Changes Over Time by Level of Risk

High risk. Significant changes over time for children classified as high risk were
found for male and female teacher-reported social competence and aggression scores; no
significant changes over time were found for self-reported prosocial behavior or aggres-
sion. These significant differences were in the hypothesized direction for social competence
and aggression. Social competence scores for high-risk children increased significantly for
boys (d = .36) and girls (d = .44). Teacher-rated aggression scores decreased significantly
for boys (d = –.13) and girls (d = –.24).

Medium risk. For children classified at medium-risk status, significant changes over
time were found for male and female teacher-reported social competence. No significant
changes over time were found for teacher-rated aggression or self-reported prosocial be-
havior and aggression. As hypothesized, medium-risk teacher-rated social competence
scores increased for boys (d = .34) and girls (d = .31).

Low risk. For children classified at low risk, significant changes were found for male
and female teacher-rated aggression. No significant changes were found for teacher-rated
social competence or self-reported prosocial behavior and aggression. Contrary to hypothe-
ses, teacher-reported aggression increased for boys (d = .31) and girls (d = .15).

Discussion

Based on criteria established by the surgeon general and the Centers for Disease Con-
trol and Prevention, the PeaceBuilders (Embry et al., 1996) violence prevention program
targets decreasing risk factors and increasing protective factors in a universal school-based
program utilizing effective strategies for behavior change (USDHHS, 2001). The current
investigation examined whether the PeaceBuilders violence prevention program had a dif-
ferential effect on children’s behavioral outcomes by levels of risk (low, medium, and
high); more specifically, we were interested in four outcomes, namely, teacher-reported
aggression and social competence and self-reported aggression and prosocial behavior. In
addition, we were interested in determining the effects of sex, age, and race on program
effectiveness. Although researchers have classified children at risk for future problems in
previous work (Lochman & The Conduct Problems Prevention Research Group, 1995;
Patterson et al., 1991), most of these comparisons have considered children’s behavior
differences at one point in time and not in the context of an intervention.
Findings indicated that the effects of PeaceBuilders were not universal across risk
categories. Significant behavior changes were found for children classified at high risk for
future violence at baseline. Consistent with expectations and previous research on differen-
tial effectiveness (Farrell, Meyer, & White, 2001; Flannery et al., 2003; Stoolmiller et al.,

TABLE 5
Winer Model (Step 2):
Pretest-Posttest Scores by Risk, Within-Subjects F Value,
and Significant Pairwise Comparisons

Low Risk Medium Risk High Risk
Pretest Posttest Pretest Posttest Pretest Posttest
Time Sig.
Risk Pairwise
M SD M SD M SD M SD M SD M SD Fd Comparisons
d

Males
Social competencea 4.13 .56 4.18 .67 3.55 .63 3.78 .74 2.93 .57 3.16 .71 10.96* b,c
Aggression (TRF)a,c 1.08 .18 1.15 .27 1.34 .39 1.36 .44 1.82 .51 1.75 .55 12.58* a,c
Prosocial behaviorb 1.86 .54 1.82 .55 1.84 .55 1.72 .57 1.82 .59 1.80 .56 1.65
Aggression (YSR)b,c 1.26 .35 1.32 .39 1.35 .43 1.39 .46 1.53 .53 1.52 .51 1.55
Females
Social competencea 4.44 .49 4.44 .60 3.93 .63 4.13 .65 3.29 .59 3.58 .72 25.54* b,c
Aggression (TRF)a,c 1.02 .07 1.06 .17 1.12 .22 1.14 .24 1.48 .48 1.37 .43 28.14* a,c
Prosocial behaviorb 2.17 .51 2.07 .53 2.10 .49 2.05 .50 2.13 .47 2.02 .55 0.58
Aggression (YSR)b,c 1.10 .24 1.12 .21 1.15 .29 1.17 .23 1.20 .32 1.23 .34 0.34

NOTE: a-low risk; b-medium risk; c-high risk.
a. Teacher reports of social competence and aggression are listed first, n ranges from 341 to 455 for teacher-reports (Grades K-5) and from 146 to 222 for child self-reports (Grades 3-5).
b. Child self-reports of prosocial behavior and aggression, n ranges from 125 to 194.
c. TRF = Teacher Report Form of Achenbach’s Child Behavior Checklist; YSR = Youth Self-Report Form of Achenbach’s Child Behavior Checklist (Achenbach, 1991).
d. F statistic and pairwise comparisons are significant at *p < .05.
199


2000), high-risk children showed the most significant changes over time; teacher-reported
aggression decreased, whereas teacher-rated social competence increased. Moreover, these
effects were found for boys and girls. However, no positive program effects were found for
either of the self-reported variables, prosocial behavior and aggression. For medium-risk
children, only teacher-rated social competence increased, whereas no effects were found
for teacher- or self-reported aggression. Findings for children classified at low risk showed
unexpected changes over time, namely, increases in teacher-reported aggression. At the
same time, these children maintained their relatively high levels of social competence. In
addition, though aggression increased, the levels of aggression still remained substantially
below medium- and high-risk groups, scores that would continue to result in low-risk
classification.
In conclusion, the findings from the current evaluation effort are encouraging.
Together with other recent efforts (e.g., CPPRG, 2002; Eron et al., 2002; Farrell, Meyer, &
White, 2001; Shapiro et al., 2002), large-scale universal violence prevention programs such
as PeaceBuilders show promise for changing children’s behaviors, in particular for chang-
ing risk and protective factors for future violence (cf., CPPRG, 2002). Our findings add to a
growing number of investigations that have provided evidence on differential program effi-
cacy for high-risk children and youth (e.g., CPPRG, 2002; Eron et al., 2002; Farrell, Meyer,
Kung, & Sullivan, 2001); they suggest that students classified at high risk for future behav-
ior problems significantly decreased on measures of aggression and increased on measures
of social competence. In addition to differential program effectiveness being examined via
regression methodology as shown by previous work, the current study demonstrated differ-
ential effectiveness through the multiple-gating procedure that utilized two variables for
risk classification, presence of negative and lack of positive behaviors at baseline. Future
evaluation research should continue to evaluate the effectiveness of violence prevention
programs in children who are most at risk.

Limitations of the Current Study

A number of limitations require some discussion. One important consideration is the
insider versus outsider perspective. In the current study, teachers reported more significant
behavior changes than did children’s self-reports similar to Shapiro et al. (2002). However,
in the current study, no significant results were found by risk level in the differential effec-
tiveness of PeaceBuilders for aggression or prosocial behavior. Findings by Stanger and
Lewis (1993) based on comparisons of behavior ratings between teacher, child, and parent
reports on the Child Behavior Checklist have some important implications for the current
study. They found that children generally report more problems than do teachers; they sug-
gested that one possible reason for this is that teachers rate behaviors only during school
hours, whereas children rate their behaviors across contexts. In addition, they suggested
that teachers attend to externalizing behaviors, such as aggression, differently than do chil-
dren, because these behaviors cause management problems and may be more salient for the
teachers than children. It may be that teachers attended to children’s changed behavior
within the school environment, whereas children attended to their behaviors in school and
in other contexts outside of schools with siblings or peers. Thus, the differential effective-
ness of PeaceBuilders would be limited in generalizability to the school environment.
Another issue requiring discussion is the one of quasi-experimental design. The cur-
rent study did not contain a true control condition. Due to Institutional Review Board (IRB)


requirements and practical considerations, all children received the intervention at some
point, thus creating a wait-list control condition, where one half of the students received the
intervention 1 year later. Children in this latter condition did complete data collections
prior to the intervention and may have been aware of PeaceBuilders prior to actual inter-
vention at their school. With quasi-experimental designs, the study is compromised
because program effects cannot be clearly determined; however, quasi-experimental
design emphasizes the ecological context and optimizes generalizability because programs
are evaluated as they are implemented (Henrich, Brown, & Aber, 1999). Third, without
comparing change over time among students classified in risk categories in intervention and
control conditions, we could not determine whether changes in reported behaviors were due
to regression to the mean. Therefore, one plausible explanation for our findings could
include regression to the mean.
Fourth, the current study was limited by participant attrition. The PeaceBuilders eval-
uation study was conducted in high-risk neighborhoods that experienced very high residen-
tial mobility that limited the number of students with longitudinal data (for a discussion, see
Flannery et al., 2003). A final limitation is that results may be attributable to teacher bias.
Additional analyses conducted by Belliston (2000) and Flannery et al. (2003) have docu-
mented varied fidelity of implementation. However, analyses have indicated that fidelity of
implementation did not affect the differential effectiveness of the program by risk category.

Conclusions/Implications

Universal, school-based programs such as PeaceBuilders show promise for reducing
aggression and increasing social competence (Flannery et al., 2003; Shapiro et al., 2002).
Such relatively low-cost programs that attempt to blanket the school population have
important policy implications in that spending only a few hundred dollars per child during
elementary school might save the criminal justice system millions later on, when individu-
als enter it during adolescence and adulthood (Cohen, 1998). Specifically, Cohen (1998)
estimated the costs of a criminal on society based on calculations such as mean number of
offenses, victim cost of crime, cost of investigation and adjudication, incarceration, fore-
gone earnings, and opportunity cost of time. He noted that the benefits of programs that
reduce crime might exceed the cost estimates computed, in terms of affecting large social
problems, reducing fear of crime, reducing private security measures, or changing lifestyle
due to decreased risk of victimization (e.g., walking vs. taking a cab). Cohen estimated that,
for a juvenile career, the present lifetime costs range between U.S.$80,000 and $325,000;
for an adult offender, $1.2 million, total costs ranging from $1.3 to $1.5 million for juvenile
and adult career offenses. When combining comorbid problems of criminality, drug use,
and high school dropout, costs to society range from $1.7 to $2.3 million (Cohen, 1998). In
contrast, the entire PeaceBuilders project budget for project administration, project devel-
opment, project implementation, training, follow-ups, evaluation design, and data collec-
tion and analysis cost less than $200 per child over the project’s 3-year period. Thus, the
cost of the program is minimal compared to potential costs due to a life of crime and vio-
lence. Universal programs such as PeaceBuilders seem effective and cost-efficient because
they can reach an entire population of children, not only children at risk. By reaching a
greater number of children, such programs change the school climate, reduce the number of
classroom disruptions, and ultimately reduce the total number of children at risk for future
violence.

Major PeaceBuilders Research Program Articles

Major PeaceBuilders Research Program Articles

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (17)

En vedette

En vedette (6)

Similaire à Major PeaceBuilders Research Program Articles

Similaire à Major PeaceBuilders Research Program Articles (20)

Plus de Dennis Embry

Plus de Dennis Embry (20)

Major PeaceBuilders Research Program Articles